نوع مقاله : مقاله پژوهشی
نویسندگان
1 دانشجوی دکترای سازههای آبی، گروه مهندسی آب، دانشگاه علوم کشاورزی و متابع طبیعی گرگان، گرگان، ایران
2 دانشیار گروه مهندسی آب دانشکده آب و خاک دانشگاه علوم کشاورزی و منابع طبیعی گرگان
3 استاد گروه مهندسی آب، دانشگاه علوم کشاورزی و منابع طبیعی گرگان، گرگان، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Monitoring the morphological changes of rivers (especially narrow-width rivers) has always been challenging. In this study, after preparing and preprocessing Sentinel-2 satellite images, the performance of two machine learning methods—supervised Support Vector Machine (SVM) classification and unsupervised K-means clustering—for extracting the centerline of the Atrak River in Golestan province was evaluated. The spectral indices NDWI, MNDWI, and AWEIsh, along with spectral bands, were used to train the models. The accuracy of the SVM model was assessed using the Kappa coefficient and IoU metrics. The river centerlines for both methods were extracted using QGIS software, and the accuracy of the results was evaluated based on RMSE, percentage length difference, and spatial agreement (using a 10-meter buffer zone). Finally, the centerline extracted by the superior model was used to calculate geometrical parameters and meander migration rates. The accuracy assessment results clearly demonstrated the superiority of the SVM method. For this method, the Overall Accuracy, Kappa coefficient, and IoU for 2016 were 96.7%, 0.9333, and 0.9354, respectively, and for 2021 were 95%, 0.9, and 0.9045, respectively. Furthermore, the RMSE of the SVM method (3.82 m and 3.35 m for 2016 and 2021, respectively) was significantly lower than that of the K-means method (5.11 m and 4.58 m, respectively). The analysis of morphological changes indicated a very high and varying migration rate of the Atrak River in different meanders, with the highest meander migration rate calculated as 39.7 meters per year. The dominant patterns of these changes were rotation and extension.
کلیدواژهها [English]
Rivers serve as vital arteries of the Earth, playing a decisive role in maintaining ecological balance and the socio-economic development of human communities. Monitoring the morphological changes of rivers, especially those with an average width of less than 50 meters, has consistently faced challenges due to the spatial resolution limitations of satellite imagery. These changes can pose significant threats to critical infrastructure such as bridges, power and gas transmission lines, roads, and urban facilities. This research aimed to provide a precise algorithm for river extraction and analysis of the morphological changes in four meanders of the Atrak River, within the vicinity of Kurand and Hoton villages in Golestan Province, over a five-year period (2016-2021). This was achieved through the integration of Sentinel-2 satellite imagery and machine learning models.
In this study, two satellite images from the dry season (July 2016 and August 2021) were utilized to minimize the effects of cloud cover and riparian vegetation. Following image pre-processing, the performance of two machine learning methods—the supervised classification Support Vector Machine (SVM) and the unsupervised K-means clustering—was evaluated. The hyperparameters of the SVM model were tuned using a Bayesian optimization algorithm. Spectral indices, including the Normalized Difference Water Index (NDWI), Modified Normalized Difference Water Index (MNDWI), and Automated Water Extraction Index (AWEIsh), along with the spectral bands, were used to train the models. The accuracy of the SVM model was assessed using the Kappa coefficient and Intersection over Union (IoU) metrics. The river centerlines for both methods were extracted using QGIS software, and their accuracy was evaluated using Root Mean Square Error (RMSE), length difference percentage, and spatialagreement (within a 10-meter buffer zone). Finally, the centerline extracted by the superior selected model (SVM) was used to calculate geomorphological parameters and the meander migration rate.
The accuracy assessment results clearly demonstrated the superiority of the SVM method. For this method, the overall accuracy, Kappa coefficient, and IoU for the years 2016 and 2021 were calculated as 96.7%, 0.9333, 0.9354, and 95%, 0.9, 0.9045, respectively. Furthermore, the RMSE of the SVM method (3.82 m and 3.35 m for 2016 and 2021, respectively) was significantly lower than that of the K-means method (5.11 m and 4.58 m, respectively). The analysis of morphological changes revealed a very high and variable migration rate across the different meanders. The highest migration rate was calculated for Meander 1, equivalent to 39.7 meters per year. The dominant pattern of these changes was identified as rotation and expansion.
This study conclusively demonstrates that the integration of Sentinel-2 satellite imagery with a machine learning framework, constitutes a highly effective methodology for monitoring planform dynamics in narrow rivers. The superior performance of the developed model successfully addresses the persistent challenge of spatial resolution limitations for accurately mapping fluvial systems with widths below 50 meters. The research quantified significant morphological instability and variable migration patterns along the studied reach, revealing a highly dynamic fluvial environment.
Conceptualization, A.Z., and A.A.D.; methodology A.Z., KH.GH. and M.M.H.; software, M.M.H.; validation, A.Z., M.M.H. and KH.GH.; formal analysis, A.Z. and A.A.D.; investigation, A.A.D.; resources, M.M.H.; data curation, A.Z.; writing—original draft preparation, M.M.H.; writing—review and editing, A.Z.; visualization, KH.GH.; supervision, A.Z. and A.A.D.; project administration, A.Z. and A.A.D.; funding acquisition, A.Z. All authors have read and agreed to the published version of the manuscript.
Data available on request from the authors.
The authors extend their gratitude to Gorgan University of Agricultural Sciences and Natural Resources for its support. The data for this study were partially provided by the Regional Water Company of Golstan (Official Letter No. 36/1404/698).
The authors confirm that the study was conducted in accordance with ethical principles, and no data fabrication, falsification, plagiarism, or misconduct occurred.
The author declares no conflict of interest.
)